Nat. Hazards Earth Syst. Sci., 11, 3415–3422, 2011 www.nat-hazards-earth-syst-sci.net/11/3415/2011/ Natural Hazards doi:10.5194/nhess-11-3415-2011 and Earth © Author(s) 2011. CC Attribution 3.0 License. System Sciences

Analysis of hail damages and temperature series for peninsular

A. Saa Requejo1, R. Garc´ıa Moreno2, M. C. D´ıaz Alvarez1, F. Burgaz3, and M. Tarquis1 1CEIGRAM. E.T.S.I. Agronomos,´ Universidad Politecnica´ de , Spain 2Universida de Da Coruna,˜ Departamento de Ciencias Da Navegacion´ e Da Terra, Facultad de Ciencias, Coruna,˜ Spain 3Entidad Nacional de Seguros Agrarios (ENESA) Ministry of the Natural, Rural and Marine Environments, Madrid, Spain Received: 11 February 2011 – Revised: 12 June 2011 – Accepted: 19 October 2011 – Published: 22 December 2011

Abstract. Hail is a serious concern for agriculture on the 1 Introduction . Hailstorms affect crop yield and/or qual- ity to a degree that depends on the crop species and the phe- Hail events represent an important natural disaster associated nological time. In Europe, Spain is one of the countries that with great economic losses in several regions, mainly in Eu- experience relatively high agricultural losses related to hail- rope. In Spain, the Combined Agrarian Insurance System storms. It is of high interest to study models that can support covers hail events for most types of crop losses. Insurance calculations of the probabilities of economic losses due to coverage of hail events provides a means of economic recov- hail damage and of the tendency over time for such losses. ery for most producers, mainly in the Mediterranean region, Some studies developed in and the Netherdlands where the highest incidence of losses due to hailstorms is show that the summer mean temperature was highly corre- recorded (Llasat et al., 2009). In Spain, the indemnification lated with a yearly hail severity index developed from hail- for these losses reached 12.5 million euro in 2007. related parameters obtained for insurance purposes. Mean- Because of the nature of the hazards related to hail events, while, other studies in the USA point out that a highly sig- the State Agrarian Insurance Body (Entidad Nacional de Se- nificant correlation between both is not possible to find due guros Agrarios, ENESA), which coordinates and links ad- to high climatic variability. ministrative activities to guarantee the development of agrar- The aim of this work is to test the correlation between av- ian insurance in Spain, is interested in models that can sup- erage minimum temperatures and hail damage intensity over port further calculations of the probabilities of economic the Spanish Iberian Peninsula. With this purpose, correlation losses and of the tendency for such losses to occur over time. analyses on both variables were performed for the 47 Span- As mentioned by Lopez´ et al. (2007), the correlation of the ish provinces (as individuals and single set) and for all crops occurrence of hail events with climatic variables is a compli- and four individual crops: grapes, wheat, barley and winter cated problem because of the small area disturbed and be- grains. Suitable crop insurance data are available from 1981 cause of the short time associated with hail events. Thus, until 2007 and based on this period, temperature data were although economic losses related to hail can be very impor- obtained. tant, the information on the climatology associated with hail This study does not confirm the results previously obtained events is very limited. It must also be recognised that most of for France and the Netherlands that relate observed hail dam- the records of hailstorm events are made for insurance pur- age to the average minimum temperature. The reason for poses. this difference and the nature of the cases observed are dis- Few studies address the climatology of hail events in Eu- cussed. rope (Tuovinen et al., 2009; Webb et al., 2009; Hyvarinen¨ and Saltikoff, 2010). Likewise, predictive models of hail events are lacking and are related in most of the cases with precipitation models. The problem is complicated because most investigations of this topic necessarily in- volve hailstorm time series that are non-homogeneous and Correspondence to: A. Saa Requejo non-continuous. For these reasons, model development ([email protected]) for different geographical regions having different climatic

Published by Copernicus Publications on behalf of the European Geosciences Union. 3416 A. Saa Requejo et al.: Analysis of hail damages and temperature series for peninsular Spain characteristics is very difficult (Piani, 2005; Fraile et al., minimum temperatures (Botzen et al. 2010). In fact, these 2003). Moreover, Piani et al. (2005) have observed that the authors found a strong positive correlation between hailstorm relationship between hailstorm events and the parameters re- events and indicators of temperature and precipitation for the lated to their occurrence seems unclear in published works Netherlands. The combination of precipitation with maxi- on the subject. mum temperatures is noted by the authors as one of the best The study of hailstorms is based on several meteorologi- indicators of hailstorm events, whereas minimum tempera- cal parameters that are involved in conditions required to pro- ture is the best indicator when temperature is used alone. duce convection (temperature, instability, a deep humid layer The authors compared the results of this model with the re- in the low and middle troposphere and an updraft to initiate sults of other studies of the problem, e.g., a study of hail- convection), as well as on the specific topographic character- storms occurring in Switzerland (Willemse, 1995). This au- istics of the geographical area studied (Dessens, 1995; Johns thor studied the statistical relationships of severe recorded and Doswell III, 1995; Siemonov and Georgiev, 2003; Lopez´ hailstorm events, based on producer damage claims recorded et al., 2007; Garc´ıa-Ortega et al., 2007). These studies based since 1949, to other climate indicators. He concluded that the their descriptive and predictive models on several parameters data could not be used to extrapolate future hailstorm trends that depended, in most of the cases studied, on the geograph- because the data have too many uncertainties and because the ical area. period considered was too brief. Lopez´ et al. (2007) demonstrated that a reliable hailstorm According to Changnon and Bigley (2005), the correlation study must include an adequate selection of meteorological of temperatures and hailstorm events is highly variable. In- variables that must depend on the specific geographical areas deed, this correlation can have either negative or positive val- involved. Also, a model that proves valuable for a specific ues. In countries where the spatial climatic variation is high geographical area may not give satisfactory results for other (the USA, for example), the value obtained for the correlation areas (Lynn et al., 2004; Chagnon and Bigley, 2005). In- may differ for different states. The same author studied the deed, Lopez´ et al. (2007) obtained varying results when they statistical relation between the mean temperatures across the attempted to apply models having high accuracy for Europe USA and hail and thunderstorm frequencies on a daily basis to other geographical areas. Some studies of hailstorm pre- (Changnon and Changnon, 2000, 2001). The results based diction have shown that instability indices by themselves are on simple correlations for different states showed that the unable to produce successful predictions of the incidence of parameters were negative or positively correlated, depending hailstorms (Manzato, 2003; Siedlecki, 2008). The predic- on the state. In general, states in the High Plains had pos- tions will depend strongly on the specific geographical area itively correlations and greater variability with time, mean- considered because the characteristics of this area will define while states near the Great Lakes exhibited negative correla- the values and parameters associated with the particular me- tions and less variability with time. In this sense, the relation- teorological conditions that occur (Kunz, 2007; Sanchez´ et ship between the indicators seemed to depend on indicators al., 2009; Dimitrova et al., 2009) related to geographical location. Alesandrov and Hoogenboom (2001) demonstrated great McMaster (1999) evaluated the magnitude of hail damage variability in climatology at the local level for Geor- to winter crops in Australia as the result of climate change gia (USA). Uncertainties about parameters are not the only in two geographical areas of New South Wales. The author problem. Additional uncertainties arise because of climate used upper-air climatic data from a global climatic model to change and associated phenomena (Soutworth et al., 2002; obtain forecasts of hail damage. For both geographical ar- Reddy et al., 2002; Tao et al., 2004) and because of social eas studied, the author found no change in hail-related crop perception of climatology in time (Liverman and Merideth, damage losses in relation to the changes in climatic indica- 2002; Camuffo and Sturaro, 2001). tors that resulted from climatic change. The situation is even more complicated because most hail- Mills (2005) mentioned that the losses due to damage storm studies are developed for insurance purposes, and caused by natural disasters have increased since 1960. This only a few of these studies are published. For example, increase has been produced in part by socioeconomic and Dessens (1995) has found that an increase in night-time tem- demographic trends. These trends may be considered when peratures could have been related to an increase in the oc- forecasting extreme climate events, mainly for insurance pur- currence of severe storms in France over the period 1946– poses. Specifically, Mills et al. (2002) found a positively re- 1992. This study shows that the summer mean temperature lation between lightning related insurance losses and average was highly correlated with a yearly hail severity index de- temperature in the USA. veloped from hail-related parameters obtained for insurance Selected parameters that are valuable for predicting hail purposes. Recently Berthet et al. (2011) found significant events in one specific site would not necessarily produce suc- correlations between physical hail parameters recorded with cessful results in others as there are many factors influenc- hailpad network in France and some monthly temperatures. ing it. In the case of Spain, climatology and topography Guided by this model, other authors found the same varies greatly among different geographical regions. Val- type of positive relationship between hail occurrence and leys, mountain ranges and plateaus generate a macroclimatic

Nat. Hazards Earth Syst. Sci., 11, 3415–3422, 2011 www.nat-hazards-earth-syst-sci.net/11/3415/2011/

Figure 1. Average hail claims per ha in 2005-2007 period for the provinces of PeninsularA. Saa RequejoSpain. et al.: Analysis of hail damages and temperature series for peninsular Spain 3417

2.1 Insurance data

Hail insurance data from ENESA between 1981 and 2007 were considered to represent good estimates of hail damage intensity because crop insurance is widely implemented in Spain. Hail insurance covers the production of 79 % of win- ter cereals, 78 % of fruit crops, 39 % of vineyard and 14 % of olive trees, with a year average of 36 901 hail claims in the period 2005–2007. The data are expressed in terms of a ratio between damage and production value, the loss-to-risk ratio for all crops insured against hail. Data were disaggregated by province. In Fig. 1 we can observe the great variation in hail claims per ha (or 10000 m2) due to climatology and crop insurance characteristics. However, data disaggregated

by month could not be obtained. We thus obtained a separate series of 27 points (1981–2007) for each of 47 provinces as Fig. 1. Average hail claims per ha in 2005–2007 period for the provinces of Peninsular Spain. well as for the whole Peninsular Spain. Five loss-to-risk ratio series were collected. The first se- ries included all insured crops. Four similar series were col- diversity that can be perceived in Fig. 1 as hail claims vari- lected for grapes, wheat, barley and winter grains (including ability for Peninsular Spain. Therefore, models of hailstorms wheat and barley). Series for other crops could not be consid- must be defined on a case-by-case basis because orography ered, owing to historical changes in the standards for damage and proximity to the sea may have only limited utility when evaluation for insurance purposes by expert witnesses. These predictive models are required for applications. changes could generate non homogeneous data series. In this context, CEIGRAM researchers tried to apply a re- liable model to relate hailstorm damage to summer temper- 2.2 Climatologic data ature across the different regions of Spain. This research aimed, for insurance purposes, to contrast results that de- The climatologic data consisted of monthly average mini- pended on specific parameters for each specific geographic mum temperatures recorded by the Agencia Estatal de Me- space. For the analysis of this problem, the model devel- teorolog´ıa (AEMet, the Meteorological State Agency) and oped by Dessens (1995) seemed appropriate for application provided for this study through ENESA. to different Spanish regions. In this particular case, provinces The weather stations were selected based on a grid approx- were the regions considered. For this purpose, the rela- imately 40 × 40 km in size. This selection was designed to tionship between average summer minimum temperature and ensure the availability of recent data, the fewest possible gaps hail severity index was studied for each growing season for in the data and the longest possible series. Several stations in different Spanish provinces. This trend was evaluated using each county were also selected to represent high- and low- separate, spatially-specific models. The aim of the study was altitude areas. The locations of the 404 stations selected are to try to confirm the link found in France and the Netherlands shown in Fig. 2. between hail tendency and temperatures and to use the results of the investigation for insurance planning purposes. In addi- 2.3 Correlations tion, authors have evaluated the model applied by this study to assess the specific damages to particular crops. This eval- Several data transformations are require to relate loss-to- uation sought to identify particular characteristics specific to risk hail damage (R) and summer minimum temperatures different crops within a growing season. (Dessens, 1995). Because R is highly variable, a logarith- mic transformation was used to normalise data (log(1+R)). Also, because the value of R can be less than 1 or even zero, 2 Data sets and methods we added an increment of 1 to R to avoid negative or inde- terminate values. The transformed hail damage index used This study covered Peninsular Spain. The subareas consid- in our analyses was therefore calculated as the logarithm of ered were the provinces shown in Fig. 1. The Spanish islands (1+R). were not considered in this work. This selection has a politi- We constructed regional time series of summer mini- cal component because selection criteria are imposed by the mum temperatures for each province by averaging the val- availability of hail damage data, and such data are referenced ues for all the weather stations in the province. The av- to these areas. We consider 47 provinces of Peninsular Spain, erage monthly minimum temperatures of June, July and each having an average extent of 10 000 km2. August were used to compute average summer minimum www.nat-hazards-earth-syst-sci.net/11/3415/2011/ Nat. Hazards Earth Syst. Sci., 11, 3415–3422, 2011

Figure 3. Correlations estimated between hail damage and minimum summer Figure 2. Map of Spain with the 47 provinces considered. Points representtemperature for Huelva and Zamora. Graphs show minimum summer meteorological stations. temperature at the x-axis (summer t min) and transformed hail damage index 3418 A. Saa Requejo et al.: Analysis of(log(1+R)) hail damages at the y-axis. and temperature series for peninsular Spain

Zamora

Huelva

Zamora

0.800 y = -0.0895x + 1.2017 0.700 R2 = 0.1927

_ 0.600 0.500 0.400 0.300 log (1+R) 0.200 0.100 0.000 Fig. 2. Map of Spain with the 47 provinces considered. Points 9.00 10.00 11.00 12.00 13.00 14.00 represent meteorological stations. summer t min

temperature, as proposed by Dessens (1995). We also com-Huelva

pute, as this author, the parameters for Peninsular Spain as a single set. 1.000 y = 0.1516x - 2.4074 Linear correlation obtained as a result of least squares fit- 0.800 2 _ R = 0.195 ting was used to relate year-to-year hail damage to summer 0.600 minimum temperature. Also, a non-parametric Kendal’s tau 0.400

is computed. (1+R) log Fig.0.200 3. Correlations estimated between hail damage and minimum summer temperature for Huelva and Zamora. Graphs show mini- 0.000 t 3 Results and discussion mum15.00 summer 15.50 temperature 16.00 at16.50 the x-axis 17.00 (summer 17.50 18.00min) and 18.50 trans- 19.00 formed hail damage index (log(1+summerR)) att min the y-axis. Correlations parameters for all 47 provinces and the whole Peninsular Spain with data for the agricultural seasons from 1981 to 2007 are presented in Table 1. Only two 5 % signif- Considering Kendal’s tau, the previous relations are con- icant Pearson correlation coefficients (see Fig. 3) are found. firmed (for Huelva and Zamora) and three new provinces These significant correlations resulted for Huelva province presented 5 % significant index (see Fig. 5). Malaga´ is a (in the southwest) and Zamora province (in the central north- province with low frequency of hail claims and the data west). The proportion of total variance explained is 19 %. (Fig. 5) shows a high frequency of low damage ratio, with ´ This value is slightly less than the value of 27 % observed a descent slope very close to zero. Alava and La Rioja are by Dessens (1995) and less than the value of 34 % reported two provinces with medium high and high hail claims per by Botzen et al. (2009), although these authors also found a ha, that are below a classical hail damage area along with variance of 18 % in their model for within-greenhouse crops. Navarra, a province with one of the highest hail claims lo- However, the main difference between the results of our cated east of them. The hail claims tendency is to increase in study and those obtained by these previous studies is that both and it became significant at 5 % according to Pearson’s the previous authors cited found a positive relationship be- r when the most extreme case is deleted. When the analysis tween hail damage and summer minimum temperatures. In is applied using Kendall’s tau, that only considers arranging contrast, our study found a similar positive relationship for the cases. The estimation is less sensitive to this extreme Huelva and an inverse (negative) relationship for Zamora, case and therefore the tendency is significant without having whereas the remaining 45 areas showed no significant cor- to remove it. relation. We also must remark that the provinces with sig- As regards the whole Peninsular Spain, we also found no nificant relationships are categorised as low hail incidence significant linear correlation with both methods (Fig. 4). provinces (see Fig. 1). A comment about outliers is relevant here. We also com- puted correlation coefficients for all areas by deleting one, two or three years data, including all the possible variations.

Nat. Hazards Earth Syst. Sci., 11, 3415–3422, 2011 www.nat-hazards-earth-syst-sci.net/11/3415/2011/ A. Saa Requejo et al.: Analysis of hail damages and temperature series for peninsular Spain 3419

Table 1. Lineal correlations statistical parameters: Pearson r and Kendall tau. Forty-seven provinces and the Peninsular Spain as a single set (PEN. SPAIN) are shown. (SoS = Sum of squares; n = number of data; F = F test parameter; pF = probability of F ; r2 = correlation coef- ficient; K-tau = kendall’s tau parameter; pK-tau = probability of K-tau). pF bold numbers correspond to 5 % significant Pearson correlation coefficients. pK-tau bold numbers correspond to 5 % significant index.

AREA SoS Regression SoS Residual n F pF r2 K-tau pK-tau ALAVA 0.726 0.647 27 3.045 0.093 0.109 0.271 0.024 ALBACETE 0.675 0.637 27 1.485 0.234 0.056 −0.140 0.153 1.612 1.455 27 2.683 0.114 0.097 −0.180 0.095 ALMERIA 0.674 0.629 25 1.802 0.192 0.067 0.100 0.242 ASTURIAS 2.438 2.416 9 0.145 0.708 0.009 0.167 0.266 AVILA 0.872 0.869 27 0.088 0.770 0.003 −0.054 0.346 BADAJOZ 1.236 1.148 27 1.910 0.179 0.071 0.128 0.174 0.802 0.702 27 3.576 0.070 0.125 −0.214 0.059 BURGOS 1.406 1.406 27 0.004 0.951 0.000 0.117 0.196 CACERES 1.462 1.458 27 0.057 0.814 0.002 −0.014 0.459 CADIZ 0.006 0.006 11 0.297 0.591 0.012 0.346 0.070 CANTABRIA 6.239 5.709 15 2.321 0.140 0.085 0.067 0.365 CASTELLON 2.318 2.318 27 0.002 0.961 0.000 −0.009 0.475 CIUDAD REAL 1.351 1.351 27 0.005 0.946 0.000 0.003 0.492 CORDOBA 0.079 0.079 27 0.107 0.746 0.004 −0.100 0.233 CUENCA 0.575 0.574 27 0.010 0.923 0.000 −0.031 0.409 GERONA 2.392 2.376 27 0.173 0.681 0.007 −0.066 0.316 GRANADA 0.868 0.866 27 0.066 0.800 0.003 0.100 0.233 GUADALAJARA 0.784 0.769 27 0.488 0.491 0.019 0.106 0.220 GUIPUZCOA 3.180 3.172 9 0.039 0.845 0.003 0.169 0.263 HUELVA 1.700 1.368 23 6.057 0.021 0.195 0.328 0.014 HUESCA 1.133 1.107 27 0.586 0.451 0.023 0.180 0.095 JAEN 0.848 0.840 27 0.229 0.636 0.009 −0.049 0.361 LA CORUNA˜ 3.656 3.635 3 0.100 0.756 0.006 0.000 0.500 LA RIOJA 1.666 1.492 27 2.914 0.100 0.104 0.237 0.042 LEON 0.961 0.947 27 0.388 0.539 0.015 −0.003 0.492 LERIDA 2.703 2.423 27 2.883 0.102 0.103 0.214 0.059 LUGO 4.523 4.300 19 1.297 0.266 0.049 −0.099 0.276 MADRID 1.055 1.040 27 0.364 0.552 0.014 −0.003 0.492 MALAGA 0.018 0.017 20 0.740 0.398 0.029 −0.295 0.035 1.118 1.094 27 0.557 0.463 0.022 −0.117 0.196 NAVARRA 1.018 0.953 27 1.695 0.205 0.063 0.168 0.109 ORENSE 1.882 1.766 22 1.637 0.213 0.061 0.022 0.444 PALENCIA 0.986 0.983 27 0.078 0.782 0.003 0.026 0.426 PONTEVEDRA 1.260 1.220 17 0.787 0.384 0.032 0.250 0.081 SALAMANCA 0.727 0.718 27 0.303 0.587 0.012 0.049 0.361 SEGOVIA 0.971 0.971 27 0.004 0.949 0.000 0.100 0.232 SEVILLA 0.405 0.380 27 1.691 0.205 0.063 0.145 0.144 SORIA 1.060 1.058 27 0.047 0.830 0.002 −0.060 0.331 TARRAGONA 1.570 1.570 27 0.002 0.962 0.000 0.088 0.259 0.738 0.736 27 0.067 0.798 0.003 −0.031 0.409 TOLEDO 1.021 1.006 27 0.371 0.548 0.015 0.049 0.361 1.661 1.566 27 1.519 0.229 0.057 0.191 0.081 VALLADOLID 0.941 0.886 27 1.546 0.225 0.058 −0.157 0.126 VIZCAYA 4.732 4.419 13 1.489 0.236 0.066 0.181 0.195 ZAMORA 0.670 0.541 27 5.967 0.022 0.193 −0.283 0.019 1.091 1.080 27 0.247 0.623 0.010 0.128 0.174 PEN.SPAIN 0.013 10.673 27 0.029 0.433 0.001 −0.009 0.457

www.nat-hazards-earth-syst-sci.net/11/3415/2011/ Nat. Hazards Earth Syst. Sci., 11, 3415–3422, 2011 Figure 5. Significant correlations according to Kendall’s tau test between hail damage and minimum summer temperature, excluding those provinces that gave a Pearson r significant. Figure 4. Correlation estimated between hail damage and minimum summer Graphs show minimum summer temperature at the x-axis (summer t min) and transformed hail temperature, at a 95% of significant level, for Peninsular Spain data as a single set and all crops. Graph shows minimum summer temperature at the x-axis (summer t min) and damage index (log(1+R)) at the y-axis. transformed3420 hail damage index (log(1+R)) at the y-axis. A. Saa Requejo et al.: Analysis of hail damages and temperature series for peninsular Spain

3.5 y = 0.0336x + 1.4788 R2 = 0.0012 3 Álava

2.5 La Rioja _ 2

1.5 log(1+R) log(1+R)

1

0.5

0 12.00 12.50 13.00 13.50 14.00 14.50 15.00 15.50 16.00 16.50 17.00 summer t min

Fig. 4. Correlation estimated between hail damage and minimum summer temperature, at a 95 % of significant level, for Peninsular Spain data as a single set and all crops. Graph shows minimum summer temperature at the x-axis (summer t min) and transformed Málaga hail damage index (log(1+R)) at the y-axis.

Two provinces change to 5 % significant coefficient deleting Álava 1 point, four new provinces change to significant r when 2 points are deleted, and 15 new provinces change when 3 points are deleted. Revising these changing provinces, 1.000 we found cases where the set of points showed a spheri- cal shape around the tendency and lost their thickness when y = 0.074x - 0.6572 0.800 2 some points located in the periphery were omitted. This _ R = 0.1086 causes an increase in the correlation coefficient; however, 0.600 they do not seem to be specific outliers that should be elim- inated as they are located in the context of the point set and characterize the variability in the time and space of hails. 0.400 Therefore, the relation between temperature and hail (1+R) log shows that it is relevant in hail claims for some provinces, 0.200 but not for overall Peninsular Spain or for most of Spain. We can explain these differences in terms of geographical 0.000 and topographical characteristics. Compared with Holland 11.00 11.50 12.00 12.50 13.00 13.50 14.00 14.50 15.00 and France, Spain has markedly higher variation in relief and is characterised by alternate plains, hilly and valley areas (see summer t min Cohen and Small, 1998, for the area distribution under and below 500 m). Another difference between our study and the works La Rioja cited involves the length of the data series analysed. Dessens (1995) analysed data from the years 1946–1992. Botzen et al. (2009) analysed monthly data from the years 1990–2005. We could not use data for years before 1981 Fig. 5. Significant correlations according to Kendall’s tau test be- tween hail damage and minimum summer temperature, excluding because ENESA and Agoseguro (a pool of private insurance 1.400 those provinces that gave a Pearson r significant. Graphs show y = 0.1117x - 1.0025 companies) felt that homogeneity of the data could not be minimum1.200 summer temperature at the x-axis (summer t min) and 2 _ guaranteed prior to this date. This problem is also mentioned transformed hail damage index (log(1+R)) at the y-axis. R = 0.1044 by Dessens (1995). The monthly data are only available for 1.000 the past five years. 0.800 Another aspect of the heterogeneity of the data results 0.600 from the characteristics of the contract periods and the insur- consideration of each crop could possibly improve the overall log (1+R) ance coverage. These characteristics produced annual time analysis0.400 of the relationship between hail damage and summer periods that could not be analysed effectively because they minimum0.200 temperatures. For this reason, several damage se- were too long. These considerations suggested that separate ries were constructed for each crop, based on the importance 0.000 11.00 12.00 13.00 14.00 15.00 16.00 Nat. Hazards Earth Syst. Sci., 11, 3415–3422, 2011 www.nat-hazards-earth-syst-sci.net/11/3415/2011/ summer t min

A. Saa Requejo et al.: Analysis of hail damages and temperature series for peninsular Spain 3421

shown by the models evaluated should result from climatic variability (very strong in Peninsular Spain) rather than from simple considerations of spatial amplitude. Further research is necessary considering other climatic variables such as precipitation and temperatures for different months and seasons (from annual to crop phenological peri- ods). At the same time, an improvement in meteorological and crop damage data resolution with respect to time and/or space is necessary to explore a new type of analysis for find- ing significant relationships.

4 Conclusions

The results of other studies linking hail damage to minimum summer temperature in countries outside Spain did not agree with our results for Peninsular Spain. For some areas we found significant relationships; however, not all of them are relevant regarding hail incidence. We must recognise that the results of the analyses could be affected by the series length, Fig. 6. Positive (red) and negative (blue) significant correla- by the data frequency (monthly or crop campaign), by the tions between hail damage and minimum summer temperature for crop considered or by changes in the crop damage evalua- (a) grapes, (b) wheat, (c) barley, and (d) all winter grains. tion process. All these factors have an obvious influence on of the crop in Spanish agriculture. The crops considered were the study. On the other hand, we cannot dismiss the differ- grapes, wheat, barley and all winter grains (including wheat ences that might result from the specific dynamical charac- and barley). teristics of each area and from the special characteristics of Tree fruit damage is of great relevance in the Spanish each area’s landscape and topography. Peninsula. However, damage to these crops was not con- Acknowledgements. This work was supported by ENESA under sidered because the process used for insurance damage eval- projects P080220C-1065 and P090220C-822. AEMet is thankfully uation changed midway through the period of study. acknowledged for providing necessary data for this study. Correlation analyses were performed for the 47 provinces and for the four crop types considered. The results of these Edited by: A. Gobin analyses are shown in Fig. 3. Only two or three provinces Reviewed by: M. A. Rodrigues Rodrigues and three other exhibited 5 % significant correlations between crop damage anonymous referees and temperature. These results are similar to those found for total agricultural damage. A negative 5 % significant corre- lation was found for grapes in Avila and Huesca and a pos- References itive correlation at Leon.´ A negative correlation was found Alexandrov, V. A. and Hoogenboom, G.: Climate variation and crop for wheat at Zamora and a positive correlation at Alicante. production in Georgia, USA during the twentieth century, Clim. A negative correlation was found for barley at Lerida´ and Res., 17, 33–43, 2001. Barcelona and a positive correlation at Alicante. For all win- Berthet, C., Dessens, J., and Sanchez, J. L.: Regional and yearly ter grains, a negative correlation was found at Zamora and variations of hail frequency and intensity in France, Atmos. Res., Lerida´ and a positive correlation at Alicante. 100, 391–400, 2011. The results for all the crops studied are consistent across Botzen, W. J. W., Bouwer, L. M., and van den Bergh, J. C. J. Peninsular Spain. Only a few scattered areas exhibited statis- M.: Climate change and hailstorm damage: Empirical evidence tically significant increasing or decreasing tendencies, with and implications for agriculture and insurance, Resour. Energy no clear connections evident in any case. Furthermore, only Econom., 32, 341–362, 2010. some provinces show consistent tendencies, and these ten- Camuffo, D. and Sturaro, G.: The climate of Rome and its action dencies occur in the context of crop inclusion, for example on monument decay, Clim. Res., 16, 145–155, 2001. Changnon, D. and Bigley, R.: Fluctuations in US freezing rain days, in the case of cereals in Alicante. Clim. Change, 69, 229–244, 2005. Our results are closer to the findings described by Changnon, S. A. and Changnon, D.: Long-term fluctuations in hail Changnon (2000, 2001, 2005) for the USA than to the find- incidences in the United States, J. Climate, 13, 658–664, 2000. ings from France and The Netherlands. Although the area Changnon, S. A. and Changnon, D.: Long-term fluctuations in thun- covered by our study is smaller than the area included in the derstorm activity in the United States, Clim. Change, 50, 489– study involving the USA, the lack of predictive consistency 503, 2001. www.nat-hazards-earth-syst-sci.net/11/3415/2011/ Nat. Hazards Earth Syst. Sci., 11, 3415–3422, 2011 3422 A. Saa Requejo et al.: Analysis of hail damages and temperature series for peninsular Spain

Cohen, J. E. and Small, C.: Hypsographic demography: The dis- Mills, E.: Insurance in a Climate of Change, Science, 5737, 1040– tribution of human population by altitude, Proc. Natl. Acad. Sci. 1044, 2005. USA, 95, 14009–14014, 1998. Mills, E., Lecomte, E., and Peara, A.: Insurers in the greenhouse, J. Dessens, J.: Severe convective weather in the context of a nighttime Insurance Reg., 21, 43–78, 2002. global warming, Geophys. Res. Lett., 22, 1241–1244, 1995. Piani, F., Crisci, A., De Chiara, G., Maracchi, G., and Meneguzzo, Dimitrova, T., Mitzeva, R., and Savtchenko, A.: Environmental F.: Recent trends and climatic perspectives of hailstorms fre- conditions responsible for the type of precipitation in summer quency and intensity in Tuscany and Central Italy, Nat. Hazards convective storms over Bulgaria, Atmos. Res., 93, 30–38, 2009. Earth Syst. Sci., 5, 217–224, doi:10.5194/nhess-5-217-2005, Fraile, R., Berthet, C., Dessens, J., and Sanchez,´ J. L.: Returns 2005. periods of severe hailfalls computed from hailpad data, Atmos. Reddy, K. R., Doma, P. R., Mearns, L. O., Boone, M. Y. L., Hodges, Res., 67–68, 189–202, 2003. H. F., Richardson, A. G., and Kakani, V. G.: Simulating the im- Garc´ıa-Ortega, E., Fita, L., Romero, R., Lopez,´ L., Ramis, C., and pacts of climate change on cotton production in the Mississippi Sanchez,´ J.L.: Numerical simulation and sensitivity study of a Delta, Clim. Res., 22, 271–281, 2002. severe hailstorm in northeast Spain, Atmos. Res., 83, 225–241, Sanchez,´ J. L., Marcos, J. L., Dessens, J., Lopez,´ L., Bustos, C., 2007. and Garc´ıa-Ortega, E.: Assessing sounding-derived parameters Hyvarinen,¨ O. and Saltikoff, E.: Social Media as a Source of Me- as storm predictors in different latitudes, Atmos. Res., 93, 446– teorological Observations, Mon. Weather Rev., 138, 3175–3184, 456, 2009. 2010. Siedlecki, M.: Selected instability indices in Europe, Theoreti. Johns, R. H. and Doswell III, C. A.: Severe local storms forecasting, Appl. Climatol., 96, 85–94, 2008. Weather Forecast, 7, 588–612, 1992. Simeonov, P. and Giorgiev, C. G.: Severe wind/hail storms over Kunz, M.: The skill of convective parameters and indices to predict Bulgaria in 1999–2001 period: synoptic- and meso-scale factors isolated and severe thunderstorms, Nat. Hazards Earth Syst. Sci., for generation, Atmos. Res., 67–68, 629–643, 2003. 7, 327–342, doi:10.5194/nhess-7-327-2007, 2007. Southworth, J., Pfeifer, R. A., Habeck, M., Randolph, J. C., Doer- Liverman, D. M. and Merideth, R.: Climate and society in the U.S. ing, O. C. and Rao, D. G.: Sensitivity of winter wheat yields in Southwest: the context for a regional assessment, Clim. Res., 21, the Midwestern United States to future changes in climate, cli- 199–218, 2002. mate variability, and CO2 fertilization, Clim. Res., 22, 73–86, Llasat, M. C., Llasat-Botija, M., Barnolas, M., Lopez,´ L., and 2002. Altava-Ortiz, V.: An analysis of the evolution of hydrom- Tao, F., Yokozawa, M., Zao, Z., Hayashi, Y., Grassl, H., and Fu, C.: eteorological extremes in newspapers: the case of Catalo- Variability in climatology and agricultural production in China in nia, 1982–2006, Nat. Hazards Earth Syst. Sci., 9, 1201–1212, association with the East Assian summer monsoon and El Nino˜ doi:10.5194/nhess-9-1201-2009, 2009. Southern Oscillation, Clim. Res., 28, 23–30, 2004. Lopez,´ L., Garc´ıa-Ortega, E., and Sanchez,´ J. L.: A short term fore- Tuovinen, J.-P., Punkka, A.-J., Rauhala, J., Hohti, H., and Schultz, cast model for hail, Atmos. Res., 83, 176–184, 2007. D. M.: Climatology of Severe Hail in Finland: 1930–2006, Mon. Lynn, B. H., Druyan, L., Hogrefe, C., Dudhia, J., Rosenzweig, C, Weather Rev., 137, 2238–2249, 2009. Goldberg, R., Rind, D., Healy, R., Rosenthal, J., and Kinney, P.: Webb, J. D. C., Elsom, D. M. and Meaden, G. T.: Severe hailstorms Sensitivity of present and future surface temperatures to precipi- in Britain and Ireland, a climatological survey and hazard assess- tation characteristics, Clim. Res., 28, 53–65, 2004. ment, Atmos. Res., 93, 587–606, 2009. Manzato, A.: A climatology of instability indices derived from Willemse, S.: A statistical Analysis and climatological interpre- Friuli Venezia Giulia soundings, using three different methods, tation of hailstorms in Switserland, Doctor of Natural Sciences Atmos. Res., 67, 417–454, 2003. Thesis Dissertation No. 11137, Swiss Federal Institute of Tech- McMaster, H. J.: The potential impact of global warming on hail nology, Zurich, 1995. losses to winter cereal crops in New South Wales, Clim. Change, 43, 455–476, 1999.

Nat. Hazards Earth Syst. Sci., 11, 3415–3422, 2011 www.nat-hazards-earth-syst-sci.net/11/3415/2011/